52981e3c53
fix: don't pass quotes to espeak
...
Previously, the text was wrapped in an additional set of quotes that was passed
to Espeak. This could result in different phonemization in certain edges and
caused the insertion of an initial separator "_" that had to be removed.
Compare:
$ espeak-ng -q -b 1 -v en-us --ipa=1 '"A"'
_ˈɐ
$ espeak-ng -q -b 1 -v en-us --ipa=1 'A'
ˈeɪ
Fixes #2619
2023-11-22 15:14:40 +01:00
c011ab7455
Update to v0.20.6
2023-11-17 15:16:32 +01:00
52cb1e2f68
Update model hash for v2.0.2
2023-11-17 15:16:32 +01:00
6075fa208c
Ensures that only GPT model is in training mode during XTTS GPT training ( #3241 )
...
* Ensures that only GPT model is in training mode during training
* Fix parallel wavegan unit test
2023-11-17 15:15:22 +01:00
a3279f9294
Make style
2023-11-17 15:15:22 +01:00
f21067a84a
Make k_diffusion optional
2023-11-17 15:15:21 +01:00
fbc18b8c34
Fix zh bug ( #3238 )
2023-11-16 17:51:37 +01:00
675f983550
Add sentence splitting ( #3227 )
...
* Add sentence spliting
* update requirements
* update default args v2
* Add spanish
* Fix return gpt_latents
* Update requirements
* Fix requirements
2023-11-16 11:01:11 +01:00
3c2d5a9e03
Remove duplicate AudioProcessor code and fix ExtractTTSpectrogram.ipynb ( #3230 )
...
* chore: remove unused argument
* refactor(audio.processor): remove duplicate stft+griffin_lim
* chore(audio.processor): remove unused compute_stft_paddings
Same function available in numpy_transforms
* refactor(audio.processor): remove duplicate db_to_amp
* refactor(audio.processor): remove duplicate amp_to_db
* refactor(audio.processor): remove duplicate linear_to_mel
* refactor(audio.processor): remove duplicate mel_to_linear
* refactor(audio.processor): remove duplicate build_mel_basis
* refactor(audio.processor): remove duplicate stft_parameters
* refactor(audio.processor): use pre-/deemphasis from numpy_transforms
* refactor(audio.processor): use rms_volume_norm from numpy_transforms
* chore(audio.processor): remove duplicate assert
Already checked in numpy_transforms.compute_f0
* refactor(audio.processor): use find_endpoint from numpy_transforms
* refactor(audio.processor): use trim_silence from numpy_transforms
* refactor(audio.processor): use volume_norm from numpy_transforms
* refactor(audio.processor): use load_wav from numpy_transforms
* fix(bin.extract_tts_spectrograms): set quantization bits
* fix(ExtractTTSpectrogram.ipynb): adapt to current TTS code
Fixes #2447 , #2574
* refactor(audio.processor): remove duplicate quantization methods
2023-11-16 10:57:06 +01:00
88630c60e5
Update to v0.20.5
2023-11-15 14:02:51 +01:00
73a5bd08c0
Fix XTTS GPT padding and inference issues ( #3216 )
...
* Fix end artifact for fine tuning models
* Bug fix on zh-cn inference
* Remove ununsed code
2023-11-15 14:02:05 +01:00
04901fb2e4
Add speed control for inference ( #3214 )
...
* Add speed control for inference
* Fix XTTS tests
* Add speed control tests
2023-11-14 16:07:17 +01:00
d96f3885d5
Update to v0.20.4
2023-11-13 17:07:25 +01:00
ac3df409a6
Merge pull request #3208 from coqui-ai/fix_max_mel_len
...
fix max generation length for XTTS
2023-11-13 14:32:56 +01:00
92fa988aec
Fixup
2023-11-13 13:44:06 +01:00
b85536b23f
fix max generation length
2023-11-13 13:18:45 +01:00
b2682d39c5
Make style
2023-11-13 13:01:01 +01:00
a16360af85
Implement chunking gpt_cond
2023-11-13 13:00:08 +01:00
6f1cba2f81
Update to v0.20.3
2023-11-09 17:41:37 +01:00
3b1e7038bc
fix(formatters): set missing root_path attribute ( #3182 )
...
Fixes #2778
2023-11-09 16:49:52 +01:00
a8e9163fb3
xtts/tokenizer: merge duplicate implementations of preprocess_text ( #3170 )
...
This was found via ruff:
> F811 Redefinition of unused `preprocess_text` from line 570
2023-11-09 16:32:12 +01:00
1b9c400bca
PyTorch 2.1 Updates (Weight Norm and TorchAudio I/O) ( #3176 )
...
* Replaced PyTorch weight_norm With parametrizations.weight_norm
* TorchAudio: Migrating The I/O Functions To Use The Dispatcher Mechanism
* Corrected Code Style
---------
Co-authored-by: Eren Gölge <erogol@hotmail.com>
2023-11-09 16:31:03 +01:00
66a1e248d0
torchaudio should use proper backend to load audio ( #3179 )
2023-11-09 16:28:39 +01:00
46d9c27212
Update to v0.20.2
2023-11-08 16:07:56 +01:00
03ad90135b
Add lang code in XTTS doc ( #3158 )
...
* Add lang code in XTTS doc
* Remove ununsed config and args
* update docs
* woops
2023-11-08 13:47:33 +01:00
78a596618a
Fix for exception on streaming if last chunk empty ( #3160 )
2023-11-08 11:32:02 +01:00
99edd6daa3
Fix ModelManager.list_models() ( #3128 )
...
* fix(utils.manage): remove hard-coded model_type variable
* refactor(utils.manage): address lint issues, fix typos
Addressed the following:
TTS/utils/manage.py:307:12: R1705: Unnecessary "else" after "return" (no-else-return)
TTS/utils/manage.py:308:21: W1514: Using open without explicitly specifying an encoding (unspecified-encoding)
TTS/utils/manage.py:299:4: R1710: Either all return statements in a function should return an expression, or none of them should. (inconsistent-return-statements)
TTS/utils/manage.py:299:4: R0201: Method could be a function (no-self-use)
TTS/utils/manage.py:314:4: R0201: Method could be a function (no-self-use)
2023-11-08 11:29:01 +01:00
77b18126c7
Merge pull request #3126 from akx/freevc-config-module
...
Move FreeVCConfig to TTS.vc.configs (like all other config classes)
2023-11-08 11:24:47 +01:00
cc6e9fcaa7
Fix #3153 ( #3169 )
2023-11-08 11:13:58 +01:00
a24ebcd8a6
Fix coqui api ( #3168 )
2023-11-08 10:51:23 +01:00
ce1a39a9a4
Add char limit warn ( #3130 )
...
* Add char limit warning
* Adding v2 langs
* cached_property for cutlet
* Fix import
2023-11-08 10:24:23 +01:00
f846a9f300
Update to v0.20.1
2023-11-07 14:17:36 +01:00
cbdbc44e0f
Fix XTTS v2.0 training recipe ( #3154 )
...
* Fix XTTS v2.0 training recipe
* Update XTTS v2 model hash
2023-11-07 14:16:44 +01:00
5f9ab6cfaa
Fix style
...
Co-authored-by: Aarni Koskela <akx@iki.fi>
2023-11-06 19:22:34 -03:00
2470599d18
Drop XTTS v1
2023-11-06 19:12:04 -03:00
13243df526
Update XTTS v1.1 files
2023-11-06 19:10:21 -03:00
09fb317e6d
Remove unused code
2023-11-06 17:36:32 -03:00
b146de4ce8
Bug fix on XTTS v2.0 Trainer
2023-11-06 20:26:01 +01:00
1b6f8d0e46
Update unit tests and recipes
2023-11-06 20:25:06 +01:00
72b2bac0f8
Load reference in 24khz to avoid issued with multiple sr references
2023-11-06 20:25:06 +01:00
00294ffdf6
Update XTTS docs
2023-11-06 20:24:06 +01:00
459ad70dc8
Add support for multiples speaker references on XTTS inference
2023-11-06 20:22:35 +01:00
f0cb19ecca
Drop diffusion from XTTS ( #3150 )
...
* Drop diffusion for XTTS
* Make style
* Drop diffusion deps in code
* Restore thrashed
2023-11-06 20:15:49 +01:00
5d418bb84a
Update docs
2023-11-06 18:48:41 +01:00
9bbf6eb8dd
Drop use_ne_hifigan
2023-11-06 18:43:38 +01:00
9d54bd7655
Fixup XTTS
2023-11-06 18:13:58 +01:00
c713a839da
Update VERSION
2023-11-06 15:51:56 +01:00
e45227d9ff
XTTS v2.0 ( #3137 )
...
* Implement most similar ref training approach
* Use non-enhanced hifigan for test samples
* Add Perceiver
* Update GPT Trainer for perceiver support
* Update XTTS docs
* Bug fix masking with XTTS perceiver
* Bug fix on gpt forward
* Bug Fix on XTTS v2.0 training
* Add XTTS v2.0 unit tests
* Add XTTS v2.0 inference unit tests
* Bug Fix on diffusion inference
* Add XTTS v2.0 training recipe
* Placeholder model entry
* Add cloning params to config
* Make prompt embedding configurable
* Make cloning configurable
* Cheap fix for a cheaper fix
* Prevent resampling
* Update model entry
* Update docs
* Update requirements
* Code linting
* Add xtts v2 to sep tests
* Bug fix on XTTS get_gpt_cond_latents
* Bug fix on rebase
* Make style
* Bug fix in Japenese tokenizer
* Add num2words to deps
* Remove unused kwarg and added num_beams=1 as default
---------
Co-authored-by: Eren G??lge <egolge@coqui.ai>
2023-11-06 14:58:18 +01:00
38f6f8f0bb
Run `make style` & re-enable it in CI ( #3127 )
2023-11-06 11:36:37 +01:00
5ae369d629
Move FreeVCConfig to TTS.vc.configs (like all other config classes)
2023-10-31 16:56:25 +02:00
Enno Hermann
Eren Gölge
Eren G??lge
Edresson Casanova
Julian Weber
Enno Hermann
Aarni Koskela
Matthew Boakes
Gorkem